multipathd/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Christophe Varoqui
   3  * Copyright (c) 2005 Kiyoshi Ueda, NEC
   4  * Copyright (c) 2005 Benjamin Marzinski, Redhat
   5  * Copyright (c) 2005 Edward Goggin, EMC
   6  */
   7 #include <unistd.h>
   8 #include <sys/stat.h>
   9 #include <libdevmapper.h>
  10 #include <wait.h>
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <fcntl.h>
  14 #include <errno.h>
  15
  16 /*
  17  * libcheckers
  18  */
  19 #include <checkers.h>
  20
  21 /*
  22  * libmultipath
  23  */
  24 #include <parser.h>
  25 #include <vector.h>
  26 #include <memory.h>
  27 #include <config.h>
  28 #include <util.h>
  29 #include <hwtable.h>
  30 #include <defaults.h>
  31 #include <structs.h>
  32 #include <callout.h>
  33 #include <blacklist.h>
  34 #include <structs_vec.h>
  35 #include <dmparser.h>
  36 #include <devmapper.h>
  37 #include <sysfs.h>
  38 #include <dict.h>
  39 #include <discovery.h>
  40 #include <debug.h>
  41 #include <propsel.h>
  42 #include <uevent.h>
  43 #include <switchgroup.h>
  44 #include <print.h>
  45 #include <configure.h>
  46
  47 #include "main.h"
  48 #include "pidfile.h"
  49 #include "uxlsnr.h"
  50 #include "uxclnt.h"
  51 #include "cli.h"
  52 #include "cli_handlers.h"
  53 #include "lock.h"
  54 #include "waiter.h"
  55
  56 #define FILE_NAME_SIZE 256
  57 #define CMDSIZE 160
  58
  59 #define LOG_MSG(a,b) \
  60         if (strlen(b)) condlog(a, "%s: %s", pp->dev, b);
  61
  62 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  63 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  64
  65 /*
  66  * global copy of vecs for use in sig handlers
  67  */
  68 struct vectors * gvecs;
  69
  70 static int
  71 need_switch_pathgroup (struct multipath * mpp, int refresh)
  72 {
  73         struct pathgroup * pgp;
  74         struct path * pp;
  75         unsigned int i, j;
  76
  77         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  78                 return 0;
  79
  80         /*
  81          * Refresh path priority values
  82          */
  83         if (refresh)
  84                 vector_foreach_slot (mpp->pg, pgp, i)
  85                         vector_foreach_slot (pgp->paths, pp, j)
  86                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  87
  88         mpp->bestpg = select_path_group(mpp);
  89
  90         if (mpp->bestpg != mpp->nextpg)
  91                 return 1;
  92
  93         return 0;
  94 }
  95
  96 static void
  97 switch_pathgroup (struct multipath * mpp)
  98 {
  99         mpp->stat_switchgroup++;
 100         dm_switchgroup(mpp->alias, mpp->bestpg);
 101         condlog(2, "%s: switch to path group #%i",
 102                  mpp->alias, mpp->bestpg);
 103 }
 104
 105 static int
 106 coalesce_maps(struct vectors *vecs, vector nmpv)
 107 {
 108         struct multipath * ompp;
 109         vector ompv = vecs->mpvec;
 110         unsigned int i;
 111         int j;
 112
 113         vector_foreach_slot (ompv, ompp, i) {
 114                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 115                         /*
 116                          * remove all current maps not allowed by the
 117                          * current configuration
 118                          */
 119                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 120                                 condlog(0, "%s: unable to flush devmap",
 121                                         ompp->alias);
 122                                 /*
 123                                  * may be just because the device is open
 124                                  */
 125                                 if (!vector_alloc_slot(nmpv))
 126                                         return 1;
 127
 128                                 vector_set_slot(nmpv, ompp);
 129                                 setup_multipath(vecs, ompp);
 130
 131                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 132                                         vector_del_slot(ompv, j);
 133
 134                                 continue;
 135                         }
 136                         else {
 137                                 dm_lib_release();
 138                                 condlog(3, "%s devmap removed", ompp->alias);
 139                         }
 140                 }
 141         }
 142         return 0;
 143 }
 144
 145 static void
 146 sync_map_state(struct multipath *mpp)
 147 {
 148         struct pathgroup *pgp;
 149         struct path *pp;
 150         unsigned int i, j;
 151
 152         vector_foreach_slot (mpp->pg, pgp, i){
 153                 vector_foreach_slot (pgp->paths, pp, j){
 154                         if (pp->state <= PATH_UNCHECKED)
 155                                 continue;
 156                         if ((pp->dmstate == PSTATE_FAILED ||
 157                              pp->dmstate == PSTATE_UNDEF) &&
 158                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 159                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 160                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 161                                   pp->dmstate == PSTATE_UNDEF) &&
 162                                  (pp->state == PATH_DOWN ||
 163                                   pp->state == PATH_SHAKY))
 164                                 dm_fail_path(mpp->alias, pp->dev_t);
 165                 }
 166         }
 167 }
 168
 169 static void
 170 sync_maps_state(vector mpvec)
 171 {
 172         unsigned int i;
 173         struct multipath *mpp;
 174
 175         vector_foreach_slot (mpvec, mpp, i)
 176                 sync_map_state(mpp);
 177 }
 178
 179 static int
 180 flush_map(struct multipath * mpp, struct vectors * vecs)
 181 {
 182         /*
 183          * clear references to this map before flushing so we can ignore
 184          * the spurious uevent we may generate with the dm_flush_map call below
 185          */
 186         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 187                 /*
 188                  * May not really be an error -- if the map was already flushed
 189                  * from the device mapper by dmsetup(8) for instance.
 190                  */
 191                 condlog(0, "%s: can't flush", mpp->alias);
 192                 return 1;
 193         }
 194         else {
 195                 dm_lib_release();
 196                 condlog(3, "%s: devmap removed", mpp->alias);
 197         }
 198
 199         orphan_paths(vecs->pathvec, mpp);
 200         remove_map(mpp, vecs, stop_waiter_thread, 1);
 201
 202         return 0;
 203 }
 204
 205 static int
 206 uev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 207 {
 208         condlog(2, "%s: add map (uevent)", dev->kernel);
 209         return ev_add_map(dev, vecs);
 210 }
 211
 212 int
 213 ev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 214 {
 215         char * alias;
 216         char *dev_t;
 217         int major, minor;
 218         char * refwwid;
 219         struct multipath * mpp;
 220         int map_present;
 221         int r = 1;
 222
 223         dev_t = sysfs_attr_get_value(dev->devpath, "dev");
 224
 225         if (!dev_t || sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 226                 return 1;
 227
 228         alias = dm_mapname(major, minor);
 229
 230         if (!alias)
 231                 return 1;
 232
 233         map_present = dm_map_present(alias);
 234
 235         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 236                 condlog(4, "%s: not a multipath map", alias);
 237                 return 0;
 238         }
 239
 240         mpp = find_mp_by_alias(vecs->mpvec, alias);
 241
 242         if (mpp) {
 243                 /*
 244                  * Not really an error -- we generate our own uevent
 245                  * if we create a multipath mapped device as a result
 246                  * of uev_add_path
 247                  */
 248                 condlog(0, "%s: devmap already registered",
 249                         dev->kernel);
 250                 return 0;
 251         }
 252
 253         /*
 254          * now we can register the map
 255          */
 256         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 257                                         start_waiter_thread))) {
 258                 sync_map_state(mpp);
 259                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 260                 return 0;
 261         }
 262         refwwid = get_refwwid(dev->kernel, DEV_DEVMAP, vecs->pathvec);
 263
 264         if (refwwid) {
 265                 r = coalesce_paths(vecs, NULL, refwwid);
 266                 dm_lib_release();
 267         }
 268
 269         if (!r)
 270                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 271         else
 272                 condlog(0, "%s: uev_add_map %s failed", alias, dev->kernel);
 273
 274         FREE(refwwid);
 275         return r;
 276 }
 277
 278 static int
 279 uev_remove_map (struct sysfs_device * dev, struct vectors * vecs)
 280 {
 281         condlog(2, "%s: remove map (uevent)", dev->kernel);
 282         return ev_remove_map(dev->kernel, vecs);
 283 }
 284
 285 int
 286 ev_remove_map (char * devname, struct vectors * vecs)
 287 {
 288         struct multipath * mpp;
 289
 290         mpp = find_mp_by_str(vecs->mpvec, devname);
 291
 292         if (!mpp) {
 293                 condlog(3, "%s: devmap not registered, can't remove",
 294                         devname);
 295                 return 0;
 296         }
 297         flush_map(mpp, vecs);
 298
 299         return 0;
 300 }
 301
 302 static int
 303 uev_umount_map (struct sysfs_device * dev, struct vectors * vecs)
 304 {
 305         struct multipath * mpp;
 306
 307         condlog(2, "%s: umount map (uevent)", dev->kernel);
 308
 309         mpp = find_mp_by_str(vecs->mpvec, dev->kernel);
 310
 311         if (!mpp)
 312                 return 0;
 313
 314         update_mpp_paths(mpp, vecs->pathvec);
 315         verify_paths(mpp, vecs, NULL);
 316
 317         if (!VECTOR_SIZE(mpp->paths))
 318                 flush_map(mpp, vecs);
 319
 320         return 0;
 321 }
 322
 323 static int
 324 uev_add_path (struct sysfs_device * dev, struct vectors * vecs)
 325 {
 326         condlog(2, "%s: add path (uevent)", dev->kernel);
 327         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 328 }
 329
 330
 331 /*
 332  * returns:
 333  * 0: added
 334  * 1: error
 335  * 2: blacklisted
 336  */
 337 int
 338 ev_add_path (char * devname, struct vectors * vecs)
 339 {
 340         struct multipath * mpp;
 341         struct path * pp;
 342         char empty_buff[WWID_SIZE] = {0};
 343
 344         pp = find_path_by_dev(vecs->pathvec, devname);
 345
 346         if (pp) {
 347                 condlog(0, "%s: spurious uevent, path already in pathvec",
 348                         devname);
 349                 if (pp->mpp)
 350                         return 0;
 351         }
 352         else {
 353                 /*
 354                  * get path vital state
 355                  */
 356                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 357                       devname, DI_ALL))) {
 358                         condlog(0, "%s: failed to store path info", devname);
 359                         return 1;
 360                 }
 361                 pp->checkint = conf->checkint;
 362         }
 363
 364         /*
 365          * need path UID to go any further
 366          */
 367         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 368                 condlog(0, "%s: failed to get path uid", devname);
 369                 return 1; /* leave path added to pathvec */
 370         }
 371         if (filter_path(conf, pp)){
 372                 int i = find_slot(vecs->pathvec, (void *)pp);
 373                 if (i != -1)
 374                         vector_del_slot(vecs->pathvec, i);
 375                 free_path(pp);
 376                 return 2;
 377         }
 378         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 379 rescan:
 380         if (mpp) {
 381                 if (adopt_paths(vecs->pathvec, mpp))
 382                         return 1; /* leave path added to pathvec */
 383
 384                 verify_paths(mpp, vecs, NULL);
 385                 mpp->action = ACT_RELOAD;
 386         }
 387         else {
 388                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 389                         mpp->action = ACT_CREATE;
 390                 else
 391                         return 1; /* leave path added to pathvec */
 392         }
 393
 394         /*
 395          * push the map to the device-mapper
 396          */
 397         if (setup_map(mpp)) {
 398                 condlog(0, "%s: failed to setup map for addition of new "
 399                         "path %s", mpp->alias, devname);
 400                 goto out;
 401         }
 402         /*
 403          * reload the map for the multipath mapped device
 404          */
 405         if (domap(mpp) <= 0) {
 406                 condlog(0, "%s: failed in domap for addition of new "
 407                         "path %s", mpp->alias, devname);
 408                 /*
 409                  * deal with asynchronous uevents :((
 410                  */
 411                 if (mpp->action == ACT_RELOAD) {
 412                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 413                         sleep(1);
 414                         update_mpp_paths(mpp, vecs->pathvec);
 415                         goto rescan;
 416                 }
 417                 else
 418                         goto out;
 419         }
 420         dm_lib_release();
 421
 422         /*
 423          * update our state from kernel regardless of create or reload
 424          */
 425         if (setup_multipath(vecs, mpp))
 426                 goto out;
 427
 428         sync_map_state(mpp);
 429
 430         if (mpp->action == ACT_CREATE &&
 431             start_waiter_thread(mpp, vecs))
 432                         goto out;
 433
 434         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 435         return 0;
 436
 437 out:
 438         remove_map(mpp, vecs, NULL, 1);
 439         return 1;
 440 }
 441
 442 static int
 443 uev_remove_path (struct sysfs_device * dev, struct vectors * vecs)
 444 {
 445         condlog(2, "%s: remove path (uevent)", dev->kernel);
 446         return ev_remove_path(dev->kernel, vecs);
 447 }
 448
 449 int
 450 ev_remove_path (char * devname, struct vectors * vecs)
 451 {
 452         struct multipath * mpp;
 453         struct path * pp;
 454         int i;
 455         int rm_path = 1;
 456
 457         pp = find_path_by_dev(vecs->pathvec, devname);
 458
 459         if (!pp) {
 460                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 461                 return 1;
 462         }
 463
 464         /*
 465          * avoid referring to the map of an orphanned path
 466          */
 467         if ((mpp = pp->mpp)) {
 468
 469                 /*
 470                  * remove the map IFF removing the last path
 471                  */
 472                 if (pathcount(mpp, PATH_WILD) > 1) {
 473                         vector rpvec = vector_alloc();
 474
 475                         /*
 476                          * transform the mp->pg vector of vectors of paths
 477                          * into a mp->params string to feed the device-mapper
 478                          */
 479                         update_mpp_paths(mpp, vecs->pathvec);
 480                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 481                                 vector_del_slot(mpp->paths, i);
 482
 483                         if (VECTOR_SIZE(mpp->paths) == 0) {
 484                                 char alias[WWID_SIZE];
 485
 486                                 /*
 487                                  * flush_map will fail if the device is open
 488                                  */
 489                                 strncpy(alias, mpp->alias, WWID_SIZE);
 490                                 if (flush_map(mpp, vecs))
 491                                         rm_path = 0;
 492                                 else
 493                                         condlog(3, "%s: removed map after removing"
 494                                                 " multiple paths", alias);
 495                         }
 496                         else {
 497                                 if (setup_map(mpp)) {
 498                                         condlog(0, "%s: failed to setup map for"
 499                                                 " removal of path %s", mpp->alias, devname);
 500                                         free_pathvec(rpvec, KEEP_PATHS);
 501                                         goto out;
 502                                 }
 503                                 /*
 504                                  * reload the map
 505                                  */
 506                                 mpp->action = ACT_RELOAD;
 507                                 if (domap(mpp) <= 0) {
 508                                         condlog(0, "%s: failed in domap for "
 509                                                 "removal of path %s",
 510                                                 mpp->alias, devname);
 511                                         /*
 512                                          * Delete path from pathvec so that
 513                                          * update_mpp_paths wont find it later
 514                                          * when/if another path is removed.
 515                                          */
 516                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 517                                                 vector_del_slot(vecs->pathvec, i);
 518                                         free_path(pp);
 519                                         return 1;
 520                                 }
 521                                 /*
 522                                  * update our state from kernel
 523                                  */
 524                                 if (setup_multipath(vecs, mpp)) {
 525                                         free_pathvec(rpvec, KEEP_PATHS);
 526                                         goto out;
 527                                 }
 528                                 sync_map_state(mpp);
 529
 530                                 condlog(3, "%s: path removed from map %s",
 531                                         devname, mpp->alias);
 532                         }
 533                         free_pathvec(rpvec, KEEP_PATHS);
 534                 }
 535                 else {
 536                         char alias[WWID_SIZE];
 537
 538                         /*
 539                          * flush_map will fail if the device is open
 540                          */
 541                         strncpy(alias, mpp->alias, WWID_SIZE);
 542                         if (flush_map(mpp, vecs))
 543                                 rm_path = 0;
 544                         else
 545                                 condlog(3, "%s: removed map", alias);
 546                 }
 547         }
 548
 549         if (rm_path) {
 550                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 551                         vector_del_slot(vecs->pathvec, i);
 552                 free_path(pp);
 553         }
 554
 555         return 0;
 556
 557 out:
 558         remove_map(mpp, vecs, stop_waiter_thread, 1);
 559         return 1;
 560 }
 561
 562 static int
 563 map_discovery (struct vectors * vecs)
 564 {
 565         struct multipath * mpp;
 566         unsigned int i;
 567
 568         if (dm_get_maps(vecs->mpvec, "multipath"))
 569                 return 1;
 570
 571         vector_foreach_slot (vecs->mpvec, mpp, i)
 572                 if (setup_multipath(vecs, mpp))
 573                         return 1;
 574
 575         return 0;
 576 }
 577
 578 int
 579 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 580 {
 581         struct vectors * vecs;
 582         int r;
 583
 584         *reply = NULL;
 585         *len = 0;
 586         vecs = (struct vectors *)trigger_data;
 587
 588         pthread_cleanup_push(cleanup_lock, vecs->lock);
 589         lock(vecs->lock);
 590
 591         r = parse_cmd(str, reply, len, vecs);
 592
 593         if (r > 0) {
 594                 *reply = STRDUP("fail\n");
 595                 *len = strlen(*reply) + 1;
 596                 r = 1;
 597         }
 598         else if (!r && *len == 0) {
 599                 *reply = STRDUP("ok\n");
 600                 *len = strlen(*reply) + 1;
 601                 r = 0;
 602         }
 603         /* else if (r < 0) leave *reply alone */
 604
 605         lock_cleanup_pop(vecs->lock);
 606         return r;
 607 }
 608
 609 static int
 610 uev_discard(char * devpath)
 611 {
 612         char a[10], b[10];
 613
 614         /*
 615          * keep only block devices, discard partitions
 616          */
 617         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 618             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 619                 condlog(4, "discard event on %s", devpath);
 620                 return 1;
 621         }
 622         return 0;
 623 }
 624
 625 int
 626 uev_trigger (struct uevent * uev, void * trigger_data)
 627 {
 628         int r = 0;
 629         struct sysfs_device *sysdev;
 630         struct vectors * vecs;
 631
 632         vecs = (struct vectors *)trigger_data;
 633
 634         if (uev_discard(uev->devpath))
 635                 return 0;
 636
 637         sysdev = sysfs_device_get(uev->devpath);
 638         lock(vecs->lock);
 639
 640         /*
 641          * device map event
 642          * Add events are ignored here as the tables
 643          * are not fully initialised then.
 644          */
 645         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 646                 if (!strncmp(uev->action, "change", 6)) {
 647                         r = uev_add_map(sysdev, vecs);
 648                         goto out;
 649                 }
 650                 if (!strncmp(uev->action, "remove", 6)) {
 651                         r = uev_remove_map(sysdev, vecs);
 652                         goto out;
 653                 }
 654                 if (!strncmp(uev->action, "umount", 6)) {
 655                         r = uev_umount_map(sysdev, vecs);
 656                         goto out;
 657                 }
 658                 goto out;
 659         }
 660
 661         /*
 662          * path add/remove event
 663          */
 664         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 665                            sysdev->kernel) > 0)
 666                 goto out;
 667
 668         if (!strncmp(uev->action, "add", 3)) {
 669                 r = uev_add_path(sysdev, vecs);
 670                 goto out;
 671         }
 672         if (!strncmp(uev->action, "remove", 6)) {
 673                 r = uev_remove_path(sysdev, vecs);
 674                 goto out;
 675         }
 676
 677 out:
 678         unlock(vecs->lock);
 679         return r;
 680 }
 681
 682 static void *
 683 ueventloop (void * ap)
 684 {
 685         if (uevent_listen(&uev_trigger, ap))
 686                 fprintf(stderr, "error starting uevent listener");
 687
 688         return NULL;
 689 }
 690
 691 static void *
 692 uxlsnrloop (void * ap)
 693 {
 694         if (cli_init())
 695                 return NULL;
 696
 697         set_handler_callback(LIST+PATHS, cli_list_paths);
 698         set_handler_callback(LIST+MAPS, cli_list_maps);
 699         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 700         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 701         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 702         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 703         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 704         set_handler_callback(LIST+CONFIG, cli_list_config);
 705         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 706         set_handler_callback(LIST+DEVICES, cli_list_devices);
 707         set_handler_callback(ADD+PATH, cli_add_path);
 708         set_handler_callback(DEL+PATH, cli_del_path);
 709         set_handler_callback(ADD+MAP, cli_add_map);
 710         set_handler_callback(DEL+MAP, cli_del_map);
 711         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 712         set_handler_callback(RECONFIGURE, cli_reconfigure);
 713         set_handler_callback(SUSPEND+MAP, cli_suspend);
 714         set_handler_callback(RESUME+MAP, cli_resume);
 715         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 716         set_handler_callback(FAIL+PATH, cli_fail);
 717
 718         uxsock_listen(&uxsock_trigger, ap);
 719
 720         return NULL;
 721 }
 722
 723 static int
 724 exit_daemon (int status)
 725 {
 726         if (status != 0)
 727                 fprintf(stderr, "bad exit status. see daemon.log\n");
 728
 729         condlog(3, "unlink pidfile");
 730         unlink(DEFAULT_PIDFILE);
 731
 732         lock(&exit_mutex);
 733         pthread_cond_signal(&exit_cond);
 734         unlock(&exit_mutex);
 735
 736         return status;
 737 }
 738
 739 static void
 740 fail_path (struct path * pp, int del_active)
 741 {
 742         if (!pp->mpp)
 743                 return;
 744
 745         condlog(2, "checker failed path %s in map %s",
 746                  pp->dev_t, pp->mpp->alias);
 747
 748         dm_fail_path(pp->mpp->alias, pp->dev_t);
 749         if (del_active)
 750                 update_queue_mode_del_path(pp->mpp);
 751 }
 752
 753 /*
 754  * caller must have locked the path list before calling that function
 755  */
 756 static void
 757 reinstate_path (struct path * pp, int add_active)
 758 {
 759         if (!pp->mpp)
 760                 return;
 761
 762         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 763                 condlog(0, "%s: reinstate failed", pp->dev_t);
 764         else {
 765                 condlog(2, "%s: reinstated", pp->dev_t);
 766                 if (add_active)
 767                         update_queue_mode_add_path(pp->mpp);
 768         }
 769 }
 770
 771 static void
 772 enable_group(struct path * pp)
 773 {
 774         struct pathgroup * pgp;
 775
 776         /*
 777          * if path is added through uev_add_path, pgindex can be unset.
 778          * next update_strings() will set it, upon map reload event.
 779          *
 780          * we can safely return here, because upon map reload, all
 781          * PG will be enabled.
 782          */
 783         if (!pp->mpp->pg || !pp->pgindex)
 784                 return;
 785
 786         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 787
 788         if (pgp->status == PGSTATE_DISABLED) {
 789                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 790                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 791         }
 792 }
 793
 794 static void
 795 mpvec_garbage_collector (struct vectors * vecs)
 796 {
 797         struct multipath * mpp;
 798         unsigned int i;
 799
 800         vector_foreach_slot (vecs->mpvec, mpp, i) {
 801                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 802                         condlog(2, "%s: remove dead map", mpp->alias);
 803                         remove_map(mpp, vecs, stop_waiter_thread, 1);
 804                         i--;
 805                 }
 806         }
 807 }
 808
 809 static void
 810 defered_failback_tick (vector mpvec)
 811 {
 812         struct multipath * mpp;
 813         unsigned int i;
 814
 815         vector_foreach_slot (mpvec, mpp, i) {
 816                 /*
 817                  * defered failback getting sooner
 818                  */
 819                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 820                         mpp->failback_tick--;
 821
 822                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 823                                 switch_pathgroup(mpp);
 824                 }
 825         }
 826 }
 827
 828 static void
 829 retry_count_tick(vector mpvec)
 830 {
 831         struct multipath *mpp;
 832         unsigned int i;
 833
 834         vector_foreach_slot (mpvec, mpp, i) {
 835                 if (mpp->retry_tick) {
 836                         mpp->stat_total_queueing_time++;
 837                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 838                         if(--mpp->retry_tick == 0) {
 839                                 dm_queue_if_no_path(mpp->alias, 0);
 840                                 condlog(2, "%s: Disable queueing", mpp->alias);
 841                         }
 842                 }
 843         }
 844 }
 845
 846 static void *
 847 checkerloop (void *ap)
 848 {
 849         struct vectors *vecs;
 850         struct path *pp;
 851         int count = 0;
 852         int newstate;
 853         unsigned int i;
 854
 855         mlockall(MCL_CURRENT | MCL_FUTURE);
 856         vecs = (struct vectors *)ap;
 857         condlog(2, "path checkers start up");
 858
 859         /*
 860          * init the path check interval
 861          */
 862         vector_foreach_slot (vecs->pathvec, pp, i) {
 863                 pp->checkint = conf->checkint;
 864         }
 865
 866         while (1) {
 867                 pthread_cleanup_push(cleanup_lock, vecs->lock);
 868                 lock(vecs->lock);
 869                 condlog(4, "tick");
 870
 871                 vector_foreach_slot (vecs->pathvec, pp, i) {
 872                         if (!pp->mpp)
 873                                 continue;
 874
 875                         if (pp->tick && --pp->tick)
 876                                 continue; /* don't check this path yet */
 877
 878                         /*
 879                          * provision a next check soonest,
 880                          * in case we exit abnormaly from here
 881                          */
 882                         pp->tick = conf->checkint;
 883
 884                         if (!checker_selected(&pp->checker)) {
 885                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 886                                 select_checker(pp);
 887                         }
 888
 889                         if (!checker_selected(&pp->checker)) {
 890                                 condlog(0, "%s: checker is not set", pp->dev);
 891                                 continue;
 892                         }
 893                         newstate = checker_check(&pp->checker);
 894
 895                         if (newstate < 0) {
 896                                 condlog(2, "%s: unusable path", pp->dev);
 897                                 pathinfo(pp, conf->hwtable, 0);
 898                                 continue;
 899                         }
 900
 901                         if (newstate != pp->state) {
 902                                 int oldstate = pp->state;
 903                                 pp->state = newstate;
 904                                 LOG_MSG(1, checker_message(&pp->checker));
 905
 906                                 /*
 907                                  * upon state change, reset the checkint
 908                                  * to the shortest delay
 909                                  */
 910                                 pp->checkint = conf->checkint;
 911
 912                                 if (newstate == PATH_DOWN ||
 913                                     newstate == PATH_SHAKY ||
 914                                     update_multipath_strings(pp->mpp,
 915                                                              vecs->pathvec)) {
 916                                         /*
 917                                          * proactively fail path in the DM
 918                                          */
 919                                         if (oldstate == PATH_UP ||
 920                                             oldstate == PATH_GHOST)
 921                                                 fail_path(pp, 1);
 922                                         else
 923                                                 fail_path(pp, 0);
 924
 925                                         /*
 926                                          * cancel scheduled failback
 927                                          */
 928                                         pp->mpp->failback_tick = 0;
 929
 930                                         pp->mpp->stat_path_failures++;
 931                                         continue;
 932                                 }
 933
 934                                 /*
 935                                  * reinstate this path
 936                                  */
 937                                 if (oldstate != PATH_UP &&
 938                                     oldstate != PATH_GHOST)
 939                                         reinstate_path(pp, 1);
 940                                 else
 941                                         reinstate_path(pp, 0);
 942
 943                                 /*
 944                                  * schedule [defered] failback
 945                                  */
 946                                 if (pp->mpp->pgfailback > 0)
 947                                         pp->mpp->failback_tick =
 948                                                 pp->mpp->pgfailback + 1;
 949                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 950                                     need_switch_pathgroup(pp->mpp, 1))
 951                                         switch_pathgroup(pp->mpp);
 952
 953                                 /*
 954                                  * if at least one path is up in a group, and
 955                                  * the group is disabled, re-enable it
 956                                  */
 957                                 if (newstate == PATH_UP)
 958                                         enable_group(pp);
 959                         }
 960                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 961                                 LOG_MSG(4, checker_message(&pp->checker));
 962                                 /*
 963                                  * double the next check delay.
 964                                  * max at conf->max_checkint
 965                                  */
 966                                 if (pp->checkint < (conf->max_checkint / 2))
 967                                         pp->checkint = 2 * pp->checkint;
 968                                 else
 969                                         pp->checkint = conf->max_checkint;
 970
 971                                 pp->tick = pp->checkint;
 972                                 condlog(4, "%s: delay next check %is",
 973                                                 pp->dev_t, pp->tick);
 974                         }
 975                         else if (newstate == PATH_DOWN)
 976                                 LOG_MSG(2, checker_message(&pp->checker));
 977
 978                         pp->state = newstate;
 979
 980                         /*
 981                          * path prio refreshing
 982                          */
 983                         condlog(4, "path prio refresh");
 984                         pathinfo(pp, conf->hwtable, DI_PRIO);
 985
 986                         if (need_switch_pathgroup(pp->mpp, 0)) {
 987                                 if (pp->mpp->pgfailback > 0 &&
 988                                     pp->mpp->failback_tick <= 0)
 989                                         pp->mpp->failback_tick =
 990                                                 pp->mpp->pgfailback + 1;
 991                                 else if (pp->mpp->pgfailback ==
 992                                                 -FAILBACK_IMMEDIATE)
 993                                         switch_pathgroup(pp->mpp);
 994                         }
 995                 }
 996                 defered_failback_tick(vecs->mpvec);
 997                 retry_count_tick(vecs->mpvec);
 998
 999                 if (count)
1000                         count--;
1001                 else {
1002                         condlog(4, "map garbage collection");
1003                         mpvec_garbage_collector(vecs);
1004                         count = MAPGCINT;
1005                 }
1006
1007                 lock_cleanup_pop(vecs->lock);
1008                 sleep(1);
1009         }
1010         return NULL;
1011 }
1012
1013 int
1014 configure (struct vectors * vecs, int start_waiters)
1015 {
1016         struct multipath * mpp;
1017         struct path * pp;
1018         vector mpvec;
1019         int i;
1020
1021         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1022                 return 1;
1023
1024         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1025                 return 1;
1026
1027         if (!(mpvec = vector_alloc()))
1028                 return 1;
1029
1030         /*
1031          * probe for current path (from sysfs) and map (from dm) sets
1032          */
1033         path_discovery(vecs->pathvec, conf, DI_ALL);
1034
1035         vector_foreach_slot (vecs->pathvec, pp, i){
1036                 if (filter_path(conf, pp)){
1037                         vector_del_slot(vecs->pathvec, i);
1038                         free_path(pp);
1039                         i--;
1040                 }
1041                 else
1042                         pp->checkint = conf->checkint;
1043         }
1044         if (map_discovery(vecs))
1045                 return 1;
1046
1047         /*
1048          * create new set of maps & push changed ones into dm
1049          */
1050         if (coalesce_paths(vecs, mpvec, NULL))
1051                 return 1;
1052
1053         /*
1054          * may need to remove some maps which are no longer relevant
1055          * e.g., due to blacklist changes in conf file
1056          */
1057         if (coalesce_maps(vecs, mpvec))
1058                 return 1;
1059
1060         dm_lib_release();
1061
1062         sync_maps_state(mpvec);
1063
1064         /*
1065          * purge dm of old maps
1066          */
1067         remove_maps(vecs, NULL);
1068
1069         /*
1070          * save new set of maps formed by considering current path state
1071          */
1072         vector_free(vecs->mpvec);
1073         vecs->mpvec = mpvec;
1074
1075         /*
1076          * start dm event waiter threads for these new maps
1077          */
1078         vector_foreach_slot(vecs->mpvec, mpp, i) {
1079                 if (setup_multipath(vecs, mpp))
1080                         return 1;
1081                 if (start_waiters)
1082                         if (start_waiter_thread(mpp, vecs))
1083                                 return 1;
1084         }
1085         return 0;
1086 }
1087
1088 int
1089 reconfigure (struct vectors * vecs)
1090 {
1091         struct config * old = conf;
1092
1093         /*
1094          * free old map and path vectors ... they use old conf state
1095          */
1096         if (VECTOR_SIZE(vecs->mpvec))
1097                 remove_maps(vecs, stop_waiter_thread);
1098
1099         if (VECTOR_SIZE(vecs->pathvec))
1100                 free_pathvec(vecs->pathvec, FREE_PATHS);
1101
1102         vecs->pathvec = NULL;
1103         conf = NULL;
1104
1105         if (load_config(DEFAULT_CONFIGFILE))
1106                 return 1;
1107
1108         conf->verbosity = old->verbosity;
1109
1110         if (!conf->checkint) {
1111                 conf->checkint = DEFAULT_CHECKINT;
1112                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1113         }
1114         configure(vecs, 1);
1115         free_config(old);
1116         return 0;
1117 }
1118
1119 static struct vectors *
1120 init_vecs (void)
1121 {
1122         struct vectors * vecs;
1123
1124         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1125
1126         if (!vecs)
1127                 return NULL;
1128
1129         vecs->lock =
1130                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1131
1132         if (!vecs->lock)
1133                 goto out;
1134
1135         pthread_mutex_init(vecs->lock, NULL);
1136
1137         return vecs;
1138
1139 out:
1140         FREE(vecs);
1141         condlog(0, "failed to init paths");
1142         return NULL;
1143 }
1144
1145 static void *
1146 signal_set(int signo, void (*func) (int))
1147 {
1148         int r;
1149         struct sigaction sig;
1150         struct sigaction osig;
1151
1152         sig.sa_handler = func;
1153         sigemptyset(&sig.sa_mask);
1154         sig.sa_flags = 0;
1155
1156         r = sigaction(signo, &sig, &osig);
1157
1158         if (r < 0)
1159                 return (SIG_ERR);
1160         else
1161                 return (osig.sa_handler);
1162 }
1163
1164 static void
1165 sighup (int sig)
1166 {
1167         condlog(2, "reconfigure (SIGHUP)");
1168
1169         lock(gvecs->lock);
1170         reconfigure(gvecs);
1171         unlock(gvecs->lock);
1172
1173 #ifdef _DEBUG_
1174         dbg_free_final(NULL);
1175 #endif
1176 }
1177
1178 static void
1179 sigend (int sig)
1180 {
1181         exit_daemon(0);
1182 }
1183
1184 static void
1185 sigusr1 (int sig)
1186 {
1187         condlog(3, "SIGUSR1 received");
1188 }
1189
1190 static void
1191 signal_init(void)
1192 {
1193         signal_set(SIGHUP, sighup);
1194         signal_set(SIGUSR1, sigusr1);
1195         signal_set(SIGINT, sigend);
1196         signal_set(SIGTERM, sigend);
1197         signal(SIGPIPE, SIG_IGN);
1198 }
1199
1200 static void
1201 setscheduler (void)
1202 {
1203         int res;
1204         static struct sched_param sched_param = {
1205                 .sched_priority = 99
1206         };
1207
1208         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1209
1210         if (res == -1)
1211                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1212         return;
1213 }
1214
1215 static void
1216 set_oom_adj (int val)
1217 {
1218         FILE *fp;
1219
1220         fp = fopen("/proc/self/oom_adj", "w");
1221
1222         if (!fp)
1223                 return;
1224
1225         fprintf(fp, "%i", val);
1226         fclose(fp);
1227 }
1228
1229 static int
1230 child (void * param)
1231 {
1232         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1233         pthread_attr_t attr;
1234         struct vectors * vecs;
1235
1236         mlockall(MCL_CURRENT | MCL_FUTURE);
1237
1238         if (logsink)
1239                 log_thread_start();
1240
1241         condlog(2, "--------start up--------");
1242         condlog(2, "read " DEFAULT_CONFIGFILE);
1243
1244         if (load_config(DEFAULT_CONFIGFILE))
1245                 exit(1);
1246
1247         setlogmask(LOG_UPTO(conf->verbosity + 3));
1248
1249         /*
1250          * fill the voids left in the config file
1251          */
1252         if (!conf->checkint) {
1253                 conf->checkint = DEFAULT_CHECKINT;
1254                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1255         }
1256
1257         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1258                 if (logsink)
1259                         log_thread_stop();
1260
1261                 exit(1);
1262         }
1263         signal_init();
1264         setscheduler();
1265         set_oom_adj(-16);
1266         vecs = gvecs = init_vecs();
1267
1268         if (!vecs)
1269                 exit(1);
1270
1271         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1272                 condlog(0, "can not find sysfs mount point");
1273                 exit(1);
1274         }
1275
1276         /*
1277          * fetch and configure both paths and multipaths
1278          */
1279         if (configure(vecs, 1)) {
1280                 condlog(0, "failure during configuration");
1281                 exit(1);
1282         }
1283
1284         /*
1285          * start threads
1286          */
1287         pthread_attr_init(&attr);
1288         pthread_attr_setstacksize(&attr, 64 * 1024);
1289         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1290
1291         pthread_create(&check_thr, &attr, checkerloop, vecs);
1292         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1293         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1294
1295         pthread_cond_wait(&exit_cond, &exit_mutex);
1296
1297         /*
1298          * exit path
1299          */
1300         lock(vecs->lock);
1301         remove_maps(vecs, stop_waiter_thread);
1302         free_pathvec(vecs->pathvec, FREE_PATHS);
1303
1304         pthread_cancel(check_thr);
1305         pthread_cancel(uevent_thr);
1306         pthread_cancel(uxlsnr_thr);
1307
1308         sysfs_cleanup();
1309
1310         free_keys(keys);
1311         keys = NULL;
1312         free_handlers(handlers);
1313         handlers = NULL;
1314         free_polls();
1315
1316         unlock(vecs->lock);
1317         pthread_mutex_destroy(vecs->lock);
1318         FREE(vecs->lock);
1319         vecs->lock = NULL;
1320         FREE(vecs);
1321         vecs = NULL;
1322         free_config(conf);
1323         conf = NULL;
1324
1325         condlog(2, "--------shut down-------");
1326
1327         if (logsink)
1328                 log_thread_stop();
1329
1330         dm_lib_release();
1331         dm_lib_exit();
1332
1333 #ifdef _DEBUG_
1334         dbg_free_final(NULL);
1335 #endif
1336
1337         exit(0);
1338 }
1339
1340 static int
1341 daemonize(void)
1342 {
1343         int pid;
1344         int in_fd, out_fd;
1345
1346         if( (pid = fork()) < 0){
1347                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1348                 return -1;
1349         }
1350         else if (pid != 0)
1351                 return pid;
1352
1353         setsid();
1354
1355         if ( (pid = fork()) < 0)
1356                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1357         else if (pid != 0)
1358                 _exit(0);
1359
1360         in_fd = open("/dev/null", O_RDONLY);
1361         if (in_fd < 0){
1362                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1363                         strerror(errno));
1364                 _exit(0);
1365         }
1366         out_fd = open("/dev/console", O_WRONLY);
1367         if (out_fd < 0){
1368                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1369                         strerror(errno));
1370                 _exit(0);
1371         }
1372
1373         close(STDIN_FILENO);
1374         dup(in_fd);
1375         close(STDOUT_FILENO);
1376         dup(out_fd);
1377         close(STDERR_FILENO);
1378         dup(out_fd);
1379
1380         close(in_fd);
1381         close(out_fd);
1382         chdir("/");
1383         umask(0);
1384         return 0;
1385 }
1386
1387 int
1388 main (int argc, char *argv[])
1389 {
1390         extern char *optarg;
1391         extern int optind;
1392         int arg;
1393         int err;
1394
1395         logsink = 1;
1396         dm_init();
1397
1398         if (getuid() != 0) {
1399                 fprintf(stderr, "need to be root\n");
1400                 exit(1);
1401         }
1402
1403         /* make sure we don't lock any path */
1404         chdir("/");
1405         umask(umask(077) | 022);
1406
1407         conf = alloc_config();
1408
1409         if (!conf)
1410                 exit(1);
1411
1412         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1413         switch(arg) {
1414                 case 'd':
1415                         logsink = 0;
1416                         //debug=1; /* ### comment me out ### */
1417                         break;
1418                 case 'v':
1419                         if (sizeof(optarg) > sizeof(char *) ||
1420                             !isdigit(optarg[0]))
1421                                 exit(1);
1422
1423                         conf->verbosity = atoi(optarg);
1424                         break;
1425                 case 'k':
1426                         uxclnt(optarg);
1427                         exit(0);
1428                 default:
1429                         ;
1430                 }
1431         }
1432
1433         if (!logsink)
1434                 err = 0;
1435         else
1436                 err = daemonize();
1437
1438         if (err < 0)
1439                 /* error */
1440                 exit(1);
1441         else if (err > 0)
1442                 /* parent dies */
1443                 exit(0);
1444         else
1445                 /* child lives */
1446                 return (child(NULL));
1447 }
1448